High-Variability urban-stream water-level and water-quality forecasting: comparative analysis of time-series deep-learning models
Water Practice & Technology · 01.07.2026 · 21 (7) · 3243-3261
DOI: 10.2166/wpt.2026.369 ↗Kurz gefasst
Fachartikel in der Zeitschrift Water Practice & Technology, online erschienen am 15.07.2026. Verfasst von Goeun Jang und 4 weiteren Autor:innen. Die Arbeit nennt 56 Literaturangaben.
Abstract (Originalsprache)
Diagram showing the study workflow: collecting Yeongsan River data, comparing four deep learning models, and selecting optimal site-dependent models for water-level and water-quality forecasting. The variability of urban-stream water level and water quality have intensified owing to rapid urbanization and climate change, necessitating prediction-based management systems for timely operational decisions. This study compared the forecasting performance of four deep-learning architectures – LSTM, GRU, CNN-LSTM, and CNN-GRU – using 10 years (2014–2023) of hourly observations from three water-level stations and four automatic water-quality stations in the Yeongsan River basin, South Korea. Model performance was evaluated at multiple lead times using Root Mean Square Error, Nash-Sutcliffe Efficiency, and the Diebold-Mariano test. For water-level prediction, GRU demonstrated optimal performance upstream, while CNN-GRU outperformed the other models at downstream sites with higher hydrological complexity; models excluding precipitation as an exogenous input consistently achieved superior short-term performance. Regarding water-quality prediction, the univariate CNN-GRU model demonstrated the best overall performance across all sites and parameters, while multivariate models reduced prediction errors for specific parameters, particularly total organic carbon and chlorophyll-a. These results highlight the importance of site- and parameter-specific model selection. Overall, this study proposes a statistically supported framework balancing predictive accuracy and computational efficiency, supporting practical water-level and water-quality management in highly variable urban streams.
Zitieren
APA
Jang, G., Oh, J., Han, H., Seo, J., & Park, Y. (2026). High-Variability urban-stream water-level and water-quality forecasting: comparative analysis of time-series deep-learning models. Water Practice & Technology, 21(7), 3243-3261. https://doi.org/10.2166/wpt.2026.369
Vancouver
Jang G, Oh J, Han H, Seo J, Park Y. High-Variability urban-stream water-level and water-quality forecasting: comparative analysis of time-series deep-learning models. Water Practice & Technology. 2026;21(7):3243-3261. doi:10.2166/wpt.2026.369
BibTeX
@article{Jang2026,
author = {Jang, Goeun and Oh, Jeongeon and Han, Hyeji and Seo, Jiwon and Park, Yong-Gyun},
title = {High-Variability urban-stream water-level and water-quality forecasting: comparative analysis of time-series deep-learning models},
journal = {Water Practice & Technology},
year = {2026},
volume = {21},
number = {7},
pages = {3243--3261},
doi = {10.2166/wpt.2026.369}
}
EUVIMED